Logit Models to Forecast Nationwide Intercity Travel Demand in the United States

Transcription

1 Logit Models to Forecast Nationwide Intercity Travel Demand in the United States Nicolas Hinze Virginia Tech Dr. Antonio Trani Virginia Tech Jeffrey K. Viken NASA Langley Samuel M. Dollyhigh ATK Nelson M. G. Guerreiro ATK Air Transportation Systems Laboratory

2 The Transportation Systems Analysis Model (TSAM) is a nationwide, multimodal model predicting intercity travel 9 million county pairs (3,076 X 3,076 counties) Automobile, commercial air, train, and air taxi travel Trips greater than 100 miles (1 Way) Business and non-business trips 5 household income groups 5 types of metropolitan statistical areas Demand forecasts to 2040 Four steps process: trip generation, trip distribution, mode choice, network assignment TSAM was initially conceived to quantify demand for NASA s Small Aircraft Transportation System (SATS) 2

3 TSAM produces detailed annual outputs for transportation planning purposes from 1995 to 2040 Automobile: Round trips between counties Vehicle Miles Traveled (VMTs) Fatalities Fuel consumption and emissions Road network link demand Commercial Airline: Round trips between counties and airports Enplanements between airports Revenue Passenger Miles (RPMs) Fuel consumption and emissions (In development) Operations and 4-D flight paths (In development) Train: Round trips between counties and stations (In development) 3

4 TSAM uses the classic four-step transportation systems modeling process Trip Generation Forecasts produced and attracted person trips by county Trip Distribution Distributes the produced trips between all counties Mode Choice Calculates the automobile, commercial airline, train, and air taxi demand between all counties Network Assignment Loads the automobile, commercial airline, train and air taxi demand onto their respective networks 4

5 Trip generation forecasts produced and attracted person trips Produced business trips in 2020 for $55K to $82K income group 5

6 Socio-economic forecasts and trip rates are combined to predict trips Trip Generation W&P 2011 Households By Income Group W&P 2011 Employment By State ATS 1995 Production Trip Rate Attraction Trip Rate ATS 1995 Trip Distribution Mode Choice W&P 2011 Households By County And Income Group Trip Generation Total Trips By Trip Purpose, Income Group and County W&P 2011 Employment By County Network Assignment * W&P 2011 = Woods & Poole Socioeconomic Database 2011 * ATS = American Travel Survey

7 Trip production rates are derived from the 1995 American Travel Survey (ATS) and household data Future produced trips are estimated using projected households per income group and trip production rates per income group Nationwide trip production rates by trip purpose, income group and MSA type are derived from the ATS trips and household counts. An iterative procedure is used to adjust the nationwide trip production rates at the State level to match the ATS published totals Business Production Trip Rate (Person Trips/Year/Household) <27K <55K <82K <137K >137K Total NYC MSA Very Large MSAs Large MSAs Medium MSAs Small/NonMSAs Total

8 Trip attraction rates are derived from the 1995 American Travel Survey (ATS) and employment data Future attracted trips are estimated using projected employment and trip attraction rates per income group Nationwide trip attraction rates by trip purpose, income group and MSA type are derived from the ATS trips and employment. An iterative procedure is used to adjust the nationwide trip attraction rates at the State level to match the ATS published totals Business Attraction Trip Rate (Person Trips/Year/Employment) <27K <55K <82K <137K >137K Total NYC MSA Very Large MSAs Large MSAs Medium MSAs Small/NonMSAs Total

9 Trip distribution assigns the produced trips to each destination county Distributed business trips in 2020 for $55K to $82K income group from Washington DC 9

10 A gravity model distributes the produced trips Trip Generation Trip Distribution Mode Choice Network Assignment T ij = trips from county i to county j, P i = trips produced from county i, A j = trips attracted to county j, ij = i jr A F j j ijr A F F ijr = the friction factor defined as an inverse function of travel distance for county ij and region r, and K ij = a socioeconomic adjustment factor for interchange ij. The trip distribution model is calibrated at the county level for the Fijs and the state level for the Kij using the ATS. The output of the trip distribution model is a set of county-tocounty trip tables by trip purpose and 5 household income groups. Region r are the following 4 types: MSA-MSA, MSA-NonMSA, NonMSA-MSA, and NonMSA-NonMSA. Each region type has different travel behavior. T P K ijr ij K ij 10

11 Mode choice calculates the automobile, commercial airline, train, and air taxi demand Automobile business trips in 2020 for $55K to $82K income group from Washington DC 11

12 The mode-choice model uses travel time and cost to split the distributed trips by mode Trip Generation Trip Distribution Mode Choice Commercial Airline Fares (DB1B) Commercial Airline Schedule (OAG) Commercial Airline Airport to Airport Travel Cost Commercial Airline Airport to Airport Travel Times Trips Between Counties For Each Trip Purpose And Income Group Mode Choice & Route Choice Driving Distances and Driving Times (Microsoft MapPoint) Train Travel Cost And Time (AMTRAK Schedule) Air Taxi Travel Cost And Time Network Assignment Trips By Mode Between Counties For Each Trip Purpose And Income Group Commercial Airline And Air Taxi Trips Between Airports For Each Trip Purpose And Income Group Train Trips Between Stations For Each Trip Purpose And Income Group 12

13 The mode choice model uses travel time and travel cost to estimate market share The mode-choice uses a BoxCox formulation for the utility λλ TTTT TTTTTTTTTTTT UU AAAAAAAA = αα AAAAAAAA 1 TTTT λλ TTTTTTTTTTTT λλ TTTT TTTTTTTTTT UU AAAAAA = αα AAAAAA 1 TTTT λλ TTTTTTTTTT λλ TTTT TTTTTTTTTTTT + αα AAAAAAAA 1 TTTT λλ TTTTTTTTTTTT λλ TTTT TTTTTTTTTT + αα AAAAAA 1 TTTT λλ TTTTTTTTTT Where: TT = Door to Door Travel Time TC = Door to Door Travel Cost αα TTTT = Travel Time Coefficient αα TTTT = Travel Cost Coefficient All lambdas are specific to TT, TC and Auto, Air. 13

14 Multiple inputs are used to estimate market share for airline routes UU ii = αα TTTT TTTT ii + αα TTTT TTTT ii + αα DDDD DDDD ii ii = RRRRRRRRRR NNNNNNNNNNNN TT = Travel Time TC = Travel Cost DS = Direct Seats From T100+ T1 α = Coefficient 14

15 Detailed county-to-county travel time and cost calculations are required for mode choice Automobile Origin County Commercial Airline Microsoft MapPoint & Congestion Delay (UMR) Driving Time And Cost Airport Access Time And Cost Processing Time At Airport Microsoft MapPoint & Congestion Delay (UMR) Schedule Delay Commercial Airline Schedule (OAG) Flight Time And Fare Processing Time At Airport Commercial Airline Fares (DB1B) & Baggage Fees Airport Egress Time And Cost Microsoft MapPoint & Congestion Delay (UMR) Federal Lodging Rates Lodging Time And Cost Lodging Time And Cost Federal Lodging Rates * UMR: Urban Mobility Report Destination County 15

16 Network assignment loads the commercial airline and air taxi demand onto the network Trip Generation Trip Distribution Mode Choice Network Assignment 16

17 The commercial airline network assignment model calculates the number of segment enplanements 17

18 Trip Generation Trip Distribution Mode Choice Road Network Assignment Commercial Airline Network Assignment Air Taxi Network Assignment Rail Network Assignment Highway Traffic Assignment (FAF/TransCAD) Commercial Airline Flight Frequencies Air Taxi Flight Frequencies Rail Frequencies Vehicle Miles Traveled (VMT) and Link Speed Airspace Simulations (ACES, RAMS) Future Feedback Loop Emissions and Fuel Consumption (Mobile 6) Highway Travel Fatalities Delays, Road Congestion Levels Delays, Airport Congestion Levels Delays, Station Congestion Levels TSAM Model Legend External Models Updated Travel Time And Cost Equilibrium? No Model Output In Development Yes - End 18

19 TSAM s ground network enables fuel consumption analysis, emission analysis, and fatality estimation The ground network module forecasts automobile link volume, speed, and time The Freight Analysis Framework (FAF) is adapted for automobile intercity traffic assignment TSAM ground network module components Network Inventory: total 93,000 links, and 87,000 nodes Traffic Analysis Zone (TAZ): 3,076 (county level) TAZ Centroid Connectors: 3,400 (from 1 to 12 connectors for each zone) Network link attributes information (Link ID, From Node and To Node ID, Miles, Number of Lanes, Functional Classification, Speed Limit) 19

20 The Freight Analysis Framework (FAF) is adapted for automobile intercity traffic assignment in TSAM 20

21 Mobile-6 is used for automobile intercity emissions estimation VMT by intercity trip vehicles Average link speed Mobile-6 Fleet by vehicle type (LDGV, LDDV) Roadway classification Other default values VMT: Vehicle Miles Traveled LDGV: Light Duty Gas Vehicle LDDV: Light Duty Diesel Vehicle NOx, HC, CO 21

22 Eurocontrol s Basic Aircraft Data (BADA) is used to estimate fuel consumption and emissions for complete flight profiles 22

23 Access to restricted geographic and demand data could help us improve TSAM Mode choice calibration of TSAM would be greatly improved if zip code/county and airport/station information from the American Travel Survey (1995) data were publicly available. This would allow us to improve our travel time and travel cost estimations. FAF-2 highway network data and truck O-D flow is crucial for highway traffic assignment, but are not open to public A 100% DB1B would be very desirable to improve the calibration of the commercial airline network assignment model and generate more accurate fares between airports. 23

24 Improved collection and distribution of future transportation survey data can greatly improve TSAM Future transportation surveys should collect and distribute: zip code/county and airport/station information travel time and travel cost information information about the alternative modes considered and their respective travel times and costs. A new American Travel Survey is needed. The current base year is Travel behavior has changed over the past 16 years. NHTS 2009 samples the travel behavior during a designated 24 hour period. ATS surveys the long distance trip behavior over 1 year. 24

25 Credits NASA Langley ATK Jeff Viken Stuart Cooke Jeremy Smith Sam Dollyhigh Nelson Guerrero Teck-Seng Kwa Virginia Tech Dr. Antonio Trani Nicolas Hinze Howard Swingle Virginia Tech Senanu Ashiabor Anand Seshadri Krishna Murthy Yue Xu DongHyeok Sohn Nola Shen Chetan Joshi Jeff Henderson Alex Vandyke Maria Rye Korea Aerospace University Dr. Hojong Baik 25